Systems
InternalScheduled agent ops

Hermes Agent Fleet

A lean fleet of 33 scheduled jobs across three machines (HQ + two workers), kept deliberately small. Monitors infrastructure, queues, backups, anomalies, outbound calling review, follow-ups, day briefs, research content. The agent layer underneath everything.

Three months in: every autonomous agent worth running needs a human checkpoint before it sends, posts, or commits. The good ones tell you exactly what they want to do and wait.
Related niche systems
System Card
categoryAgent infrastructure
statusInternal
page/open-agents
Public version

Agent operations, not agent demos

01

Scheduled jobs and triggers

02

Model/provider routing and budgets

03

Human approval checkpoints

04

Notifications, docs, and operational updates

A simplified view of the Hermes fleet. Logs, private emails, and credentials are intentionally not shown.
What it proves

Agents run as monitored operations with schedules, budgets, approvals, and human checkpoints.

Problem

Agent demos do not handle wakeups, retries, governance, logs, or approval routing.

1

What it does

Defines what each agent is allowed to do. Queues their work. Monitors their runs. Routes the parts that need a human back to me with enough context to decide in 60 seconds.

2

What's actually running

Infra health checks, queue health, database backups + anomaly checks, app security, doc maintenance, project status updates. Plus an EA-style M365 layer: inbox triage, follow-up sweep, meeting prep, day brief, and research-content pipelines.